[
https://issues.apache.org/jira/browse/FLINK-29238?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Weijie Guo updated FLINK-29238:
-------------------------------
Description: Hybrid shuffle relies on the index to read the disk data.
Since the spilled data may be consumed from memory, the readable status is
introduced. For the readable buffer, FileDataManager does not pre-load it.
However, when the downstream fails, the previous readable status will be used
incorrectly, resulting in that some buffer cannot be read correctly. (was:
Hybrid shuffle relies on the index to read the disk data. Since the spilled
data may be consumed from memory, the readable status is introduced. For the
readable buffer, FileDataManager does not do the pre-load. However, when the
downstream fails, the previous readable status will be used incorrectly,
resulting in that some buffer cannot be read correctly.)
> Wrong index information will be obtained after the downstream failover in
> hybrid full mode
> ------------------------------------------------------------------------------------------
>
> Key: FLINK-29238
> URL: https://issues.apache.org/jira/browse/FLINK-29238
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Network
> Affects Versions: 1.16.0
> Reporter: Weijie Guo
> Priority: Blocker
> Fix For: 1.16.0
>
>
> Hybrid shuffle relies on the index to read the disk data. Since the spilled
> data may be consumed from memory, the readable status is introduced. For the
> readable buffer, FileDataManager does not pre-load it. However, when the
> downstream fails, the previous readable status will be used incorrectly,
> resulting in that some buffer cannot be read correctly.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
